Changes in version 0.3.0 (2026-05-26)                  

Bug fixes

  - Fixed issue where off-diagonal elements were counted twice when
    computing the average sample size.

New defaults

  - When both mat and data are provided, mat is used for network
    estimation, while data is only used for additional calculations
    (e.g., for compatibility with bootnet).
  - cor_calc() now automatically scales raw data when no variables are
    treated as ordered. This was already the case when called
    indirectly, but now also applies when calling it directly. Only
    affects the returned means, not the estimated correlations.

New arguments and logic

  - Added means argument to regularization_net(). Not intended for
    typical use, but needed for integration with bootnet.
  - ns no longer accepts vectors in regularization_net(), instead
    accepts a matrix.
  - Added network_vars and auxiliary_vars arguments to cor_calc(),
    neighborhood_net(), and regularization_net(). These arguments accept
    both variable names and numeric indices referring to the provided
    data.

New features

  - Functions now return imputed data as a mids object when using
    stacked multiple imputation.
  - With the new network_vars and auxiliary_vars arguments, users can
    now specify which variables are used for network estimation and
    which additional variables are included for correlation estimation
    in the presence of missing data.

Refinements

  - Using only a single value for one of the regularization parameters
    in nonconvex regularization now results in a message rather than a
    warning.
  - Updated all dummy data sets, including the complete versions, and
    introduced varying amounts of missingness across variables in the
    incomplete data sets.

                 Changes in version 0.2.0 (2026-01-17)                  

Change of argument names and input

  - The argument k, which controlled the penalty term in
    information-criterion calculations, has been removed for security
    reasons. Instead, the penalty type is now specified via the argument
    ic_type (see the corresponding help pages).

New defaults

  - Changed the default handling of the ridge penalty in the
    multiple-imputation pmm workflow when looking for donors through
    regressions (mice). Instead of forcing ridge = 0, the function now
    uses the default value defined by mice, ensuring consistent and
    method-appropriate regularization.

New features

  - Added function reg_network() for network estimation using
    regularization, supporting both convex and non-convex penalties as
    well as multiple options for computing the likelihood in the
    information criterion when missing values are present.
  - Added function ordered_suggest(), a heuristic procedure for
    identifying variables that may be treated as ordered categorical
    based on their distribution and available information.
  - Added dummy data sets mantar_dummy_full_cat and
    mantar_dummy_mis_cat, containing only ordered categorical variables
    (with and without missing values).
  - Added dummy data sets mantar_dummy_full_mix and
    mantar_dummy_mis_mix, containing mixtures of ordered categorical and
    continuous variables (with and without missing values).

Improvements

  - Added support for treating variables as ordered categorical in the
    estimation of correlations.
  - Renamed dummy data sets to mantar_dummy_full_cont and
    mantar_dummy_mis_cont to better reflect that they contain only
    continuous variables.
  - Improved documentation for several functions and updated the README
    to reflect new functionality.

                 Changes in version 0.1.0 (2025-07-11)                  

  - Initial release